CS

chmod777john/swarm-ide

Deployment & DevOps

This page was translated automatically. Technical details may differ from the original. View the original

1.5K stars Quality 55 Trend 24

An IDE built for Agent Swarms, supporting Kimi-2.5, GLM-4.7, etc. (even models that are not reinforcedly learned like K2.5 can be used). An IDE built for Agent Swarms, supporting Kimi-2.5, GLM-4.

Overview

- Demo: - Detailed video: https://www.bilibili.com/video/BV1X163BQE5c/?share_source=copy_web&vd_source=e0705640ea2f51669a392fb07684e286 https://zhuanlan.zhihu.com/p/2000736341479138182 - Arbitrary dynamic creation of sub-agents - Messages can be sent to any agent - WeChat-style chat interface, intervene at any time with any sub-agent - Streaming graph dynamically displays collaboration status ~~It's worth noting that this project had already proposed the swarm model in Kimi-Swarm and Claude Team. Especially Claude Team, a close comparison reveals its main ideas (dynamic dispatch, communication between humans and arbitrary agents) and the project's design, which to some extent shows the author's vision and design have reached an advanced level. At a time when static LangGraph frameworks were mainstream, being able to independently create such a design was quite ahead of its time. I put the project whitepaper on the blockchain at that time. If you're worried I'm bragging, you can check the blockchain chain timestamp yourself.~~ - Graph directly displays swarm topology and real-time communication links - Tree-like multi-level dialogue list: allows selection of any agent conversation (even deep level) like WeChat - LLM history panel: displays the agent's context in real time, no longer a black box - real-time streaming output tool-call parameters - Minimalist Primitives: The system relies on only a small amount of communication primitives to express multi-agent behaviors (the core is create + send, from which complex collaboration is combined). - Liquid topology: Topology is not preset and self-evolves during operation; When encountering complex tasks, agents actively "hire" subordinates. - Flat collaboration: humans can intervene at any level like chatting, making complex topologies observable, debuggable, and intervenable. This system needs to run on Linux. If you don't have Linux or encounter problems during installation, you can try using the free virtual machine provided by GitHub. After clicking the link to create the virtual machine, you can execute the following commands He directly told him, "Create three sons, send them messages each, and have them create three grandsons on their own again."

Overview (machine translated)

English README

🎬 Video [Open Source Version Kimi-K2.5 Beehive Multi-Agent Agent]

Join WeChat groups

Zhihu Articles

https://zhuanlan.zhihu.com/p/2000736341479138182

Advantages

  • Create any dynamic sub-agent
  • Messages can be sent to any agent
  • WeChat-style chat interface, allowing any sub-agent to be intervened at any time
  • Streaming graph dynamically displays collaboration status

Comparison

It’s worth noting that this project had already independently proposed the swarm model before Kimi-Swarm and Claude Team. Especially Claude Team—a closer comparison reveals its main ideas (dynamic dispatch, communication between humans and arbitrary agents) align perfectly with the project’s design, which to some extent shows the author’s vision and design have reached an advanced level. Being able to independently create such a design at a time when static LangGraph frameworks were popular was quite ahead of its time. I put the project whitepaper on the blockchain back then. If you’re worried I’m bragging, you can check the blockchain chain timestamp yourself.

Comparison Items Kimi-Swarm Claude Agent Team Swarm-IDE
Supports nested Agent ❌ ❌ ✅
Supports Agent-to-Agent Communication ❌ ✅ ✅
Communication for Sub-Agents ❌ ✅ ✅
Supports group chat mode ❌ ❌ ✅
Supports visualization ❌ ❌ ✅
Open Source? ❌ ❌ ✅
Release Date 2026.1.27 2026.2.6 2026.1.2

Interface Design

  • Graph directly displays colony topology and real-time communication links
  • Tree-like multi-level dialogue list: allows you to select any agent conversation (even deep ones) just like WeChat
  • LLM history panel: Real-time display of the agent’s context; the agent is no longer a black box
  • Real-time streaming output of tool-call parameters

Philosophy

  • Minimalist Primitives: The system relies on only a small amount of communication primitives to express multiple Agent behaviors (the core is create + send, from which complex collaboration is combined).
  • Liquid topology: topology is not preset and self-evolves during operation; When encountering complex tasks, the Agent proactively “hires” subordinates.
  • Flat collaboration: humans can intervene at any level like chatting, making complex topologies observable, debuggable, and intervenable.

Concept

This page was translated automatically. Technical details may differ from the original.

Recommended Tools

Try a different keyword or remove a filter.

Install

npx skillfish add chmod777john/swarm-ide
New to MCP? Start with the mcpskill.org guide →